Stability classification of deformable mirror in adaptive optics based on logistic regression

In order to make adaptive optics systems more intelligent and unattended, a machine or an algorithm is needed to replace human operators. However, many factors can cause adaptive optics systems unstable and even damage the vulnerable surface of the deformable mirror. As the most expensive element in adaptive optics system, the safety of the deformable mirror must be ensured. The main difficulty to construct a classifier to distinguish abnormal situations is that it is hard to define the normal or abnormal situations in specific mathematical equations. Therefore, we build an adaptive optics system experimental platform with a 61-element deformable mirror to collect data in closed-loop. We then train a classifier based on logistic regression using the data we have collected. Results show that this method is effective in classifying the running states of deformable mirror, it achieves high accuracy in the training set, the test set and four validation datasets.